A common assumption that increasing demographic representation in training data is sufficient to mitigate AI bias becomes problematic when models trained in one sociotechnical context are deployed into others whose realities were not meaningfully represented during development, particularly in health-related AI where outcomes are shaped by lifestyle and environmental factors. Large-scale audits of biometric and healthcare datasets reveal widespread deficiencies in fairness, privacy, and regulatory compliance, and simulation-based studies show that achieving parity in model performance across groups does not necessarily translate into equitable outcomes. Modern machine learning systems frequently rely on spurious correlations enabling high performance within specific datasets but failing under distributional shifts, a phenomenon described as the "Clever Hans" effect, while model representations often diverge from human conceptual organization. Studies of large language models across different sociocultural contexts continue to document persistent biases shaped by underlying data distributions and societal structures, and these biases are amplified when models treat populations as discrete or homogeneous groups, ignoring the intersectional nature of identity and contextual factors.